← The Compass reference

Part Four: Buy Evidence — Chapter 7: The Smallest Useful Experiment

Real work, small boundary

Use real inputs, representative users, and a live workflow when the risk allows it. Invented data can help you demonstrate a feature, but it usually leaves important questions about the organization unanswered.

Keep the scope manageable: one product line, queue, document type, customer segment, internal team, or recurring decision may be enough. Choose the boundary for the business evidence it can produce. A tiny build that tells you nothing useful is still more work than you needed to do.